Comfyui-ZiT-Lora-loader

ZImageLoraAutoLoader

Self-contained version of the above — analysis and application in one node. model in, patched model out.

Loaders/Z Image · 1 Eingabe · 2 Parameter · 2 Ausgaben

ZImageLoraAutoStrength

Reads the LoRA's weight tensors directly and computes per-layer strengths from the actual training signal in the file. One knob: globalstrength.

Loaders/Z Image · 0 Eingaben · 2 Parameter · 4 Ausgaben

ZImageTurboLoraLoader

| Input | Type | Description | |---|---|---| | model | MODEL | Z-Image Turbo / Lumina2 model | | loraname | dropdown | LoRA file from models/loras | | strengthmodel | float | Global LoRA strength (-20.0 to 20.0) | | autoconvertqkv | boolean | Fuse separate Q/K/V to Z-Image's native fused QKV format | | loranameoverride | string (link) | Optional — overrides the dropdown when connected | | layerstrengths | string (link) | Optional — per-layer JSON from Auto Strength node | The graph widget shows 30 columns, one per transformer layer, each split between attention (purple, top) and feed-forward (teal, bottom). Drag to adjust. Shift-drag moves all active layers together. Click to toggle.

Loaders/Z Image · 1 Eingabe · 3 Parameter · 1 Ausgabe

ZImageTurboLoraStack

Apply up to 10 LoRAs in sequence with independent strength, enable toggle, and QKV fusion per slot.

Loaders/Z Image · 1 Eingabe · 0 Parameter · 1 Ausgabe